Strategic Default and Personal Credit : The Brazilian Natural Experiment

Transcription

1 Strategic Default and Personal Credit : The Brazilian Natural Experiment Christiano Arrigoni Coelho BCB and PUC-Rio Bruno Funchal y EPGE/FGV Abstract Brazil provided at 2004 an interesting natural experiment concerning personal credit. The government implemented a new law that allows some nancial institutions to o er a speci c type of credit. This new law removes a signi cant share of the moral hazard problem, since the lenders of this type of credit are able to deduct the debt repayment directly from the debtors payroll check, eliminating the choice of strategic default when debtors are able to pay their loans with their wage. Taking advantage of this fact, we estimate - using the di erence-in-di erence procedure - the impact of the new law on the interest rate and the volume of personal credit. We nd that the rst variable su ers a negative impact, while on the second one the e ect was positive, as expected, improving credit market conditions. Both e ects are highly signi cant. Keywords: Credit Market, default, moral-hazard, di erencein-di erence, law. JEL Codes: E44, D01, C33, K00. 1 Introduction At the end of 2003 the Brazilian Congress approved a new law that regulates payroll loans. The payroll loan is a type of personal credit with repayments directly deducted from the borrowers payroll check, which, in practice, makes a collateral out of future income. Before the christiano.coelho@bcb.gov.br. Mr. Coelho would like to stress that opinions expressed here are solely his, and do not re ect any o cial position of the Brazilian Central Bank. y bfunchal@fgvmail.br 1

2 implementation of the new law, only workers, pensioners and retired workers from the public sector had this type of credit available. The new law comes to provide such credit to private workers associated to trade unions, pensioners and retired workers from the National Institution of Social Security (INSS) 1. Only some nancial institutions (those authorized by the government) are able to provide this loan for the INSS bene ciaries. The reduction of the interest rate and the expansion path of new loans can be seen in the gures 1 and 2 respectively (line with triangles). Notice that the more pronounced e ect is on the called "treatment group" (line with diamonds), composed by the nancial institutions able to provide the payroll loans. The interest rate line of treatment group clearly detaches from the interest rate o ered by the control group (line with circles) at the same period that the law was implemented. The same happens with the volume of new loans, where the increase was stronger for the treatment group than for the control group after the new law. The explanation of this fact is that such loan eliminates debtors choice of no repayment when they have enough wage to do it, vanishing a signi cant part of the strategic default. The reduction of default s probability increases the expected repayment for the lenders making them willing to o er more credit at better terms. Taking advantage of this natural experiment we estimate, using the di erence in-di erence procedure, how it a ects the personal credit market. This empirical approach has some advantages. Our estimator has less bias than a simple mean estimator once we have a control group. For example, in the period before the new law s enforcement the mean of the basic interest rate (taxa selic) was 1.63% per month while in the period after the enforcement this mean rate was 1.36%. If we estimated the new law s e ect using the simple di erence between the interest rate for the treatment group before and after new law s enforcement we would have a biased estimator, since part of the interest rate decrease in this period is due to the basic interest rate fall. With the di erencein-di erence procedure we can have an unbiased estimator without any kind of macroeconomic control. Otherwise, if we estimated the e ect using the simple mean we would have to include macroeconomic controls, which increase the number of parameters to be estimated, putting more uncertainty in our estimation and reducing the power of the test. Also, the exogenous nature of the new law is a good opportunity to study the informational costs of personal credit market without endogeneity problems. 1 The Brazilian pension system, a pay-as-you-go scheme, is publicly managed by this governmental agency, INSS. 2

3 The empirical results show that the volume of new loans and the total amount of personal credit increased, while interest rate decreased although at a lower rate. It indicates that in Brazil the cost provided by information failures as moral hazard is signi cantly high for volume of credit o ered to the population but not so high for the interest rate charged for it. This could be saw as an evidence of credit rationing or market power in the bank sector. 7,00% 6,00% Figure 1: Evolution of the Interest Rate Interest rate total Interest rate treatment group Interest rate control group 5,00% 4,00% 3,00% 2,00% 1,00% 0,00% jan/03 abr/03 jul/03 out/03 jan/04 abr/04 jul/04 out/04 jan/05 abr/05 jul/05 out/05 8,00 7,00 Figure 2: Evolution of the New loans New loans total New loans treatment group New loans control group 6,00 R$ billions 5,00 4,00 3,00 2,00 1,00 0,00 jan/03 abr/03 jul/03 out/03 jan/04 abr/04 jul/04 out/04 jan/05 abr/05 jul/05 out/05 On this subject Costa and Mello (2005) estimated the e ect of judicial decision on the legality of payroll debit loans in Brazil. In June 2004, a high-level federal court upheld a regional court ruling that had 3

4 declared payroll deduction illegal. Evidence indicates that it had an adverse impact on banks risk perception, on interest rates and on the amount lent. Coelho, Garcia and Mello (2006) studied the presence of credit rationing in the Brazilian market using two di erent tests. First, they tested interest rate stickiness and then they do a natural experiment using the new law in order to detect credit rationing for corporate rms. In the rst test they found no evidence of interest rate stickiness. In the second one, interest rate went up and new loans volume went down after a negative supply shock. Both evidences go on opposite directions to credit rationing theory predictions. The remainder of the paper is structured as follows: in section 2 we describe the new law and its particularities; in section 3 we develop a model for personal credit market to support our empirical claims; in section 4 we describe the database used and the main descriptive statistics; section 5 presents the empirical results; and section 6 concludes. 2 The New Law The speci c legislation of payroll loans is not new in the Brazilian nancial scenario. The Law 8,112 of December 1990 already validated such type of credit, however applying just to workers, retires and pensioners of the public sector. For the private sector there was no speci c law so far, which brought serious di culties for the development of this type of credit. Private retires and pensioners didn t have permission to do such operation and private workers should create a particular instrument between three parts: employee, employer and nancial institution. The signi cant change in this legislation occurred in September 2003, when the government sent to the House of Representatives the MP that subsequently, in December 2003, turned into Law 10,820. The new law created juridical security of salary consignation through private companies and the National Institution of Social Security (INSS) to private sector formal workers and retired workers respectively. Such type of loans has some boundaries related to the agent income, however. Monthly deductions are limited to thirty percent of agents disposable wage 3 ; the loans should have x payment during the amortization period; and rescissory earns 4 can be consigned for the amortization of the remainder debt. Employers have several obligations relative to the values 2 MP is the abbreviation of Medida Provisória that is a legislative device in which the executive sends a bill to congress that is e ective immediately, pending approval. It has an urgency status that forces the legislator to appreciate its merit. For practical purposes, it is almost equivalent to a full-blown law. 3 Wage after the compulsory deductions like taxes, pensions, etc. 4 Rescissory earns comprise all rights of the employee for a contractual break. 4

5 and information that are passed to the nancial institutions and employees. To make competitive conditions to the employees, the participation of the trade union entities representing the employees is obligatory, but despite the agreement between the trade unions and some nancial institution, the employee is free to choose any nancial institution that supply this type of credit. Despite the law be in force since September 2003, just in April it really took e ect when the government authorized the Caixa Econômica Federal 6 to o er this type of credit to the pensioners and retired workers from INSS, supported by the new law. Subsequently, BMG (Bank of Minas Gerais) was the rst private bank to assign such agreement. Today there are forty four nancial institutions authorized by the government to supply this type of credit to pensioners and retired workers from INSS 7. 3 Theory Consider a consumer who lives for two periods and maximizes utility over her consumption c:the consumer born with some amount of durable goods of value D (like house, car, etc) that he or she consumes in both periods, but it depreciates at rate : Period 1 income w 1 is observed but the second period income is uncertain, varying according to the states of nature. To the regular workers the good states means that their wage takes a higher value, while bad states means lower values, thus w 2s 2 [w 21 ; w 2S ]: Since pensioners and retired people earns a x income, let say w, we model the uncertainty assuming that an adverse event may happen, obligating them to spend an extra g s of their income (ex: medicine), thus (w g s ) + = w 2s 2 [w 21 ; w 2S ]; where w = w 2S : Each state occurs with probability p s, where p s > 0 8s and X p s = 1: s There is a large number of agents divided in two di erent groups: borrowers and lenders. Borrowers may be thought as consumers and lenders as nancial institutions. Each lender is endowed with enough money to supply credit to consumers. Such lenders endowment may be used either to lend to a borrower with rate r; or to purchase a risky-free asset paying an exogenously given rate of return r f : If the borrowers default, the lenders can call the court to seize the 5 To de ne the exact month that a institution became able to give payroll loans we used the following criteria: when the date of the deal signature was in the rst half of the month we considered the same month, otherwise we considered the next month. For example, Caixa Econômica singed the deal in 04/19/2004, so we considered May as the date that it became able. 6 The Caixa Econômica Fedral is the biggest public bank. 7 For more details about the date of authorization see Appendix. 5

6 borrowers assets. In the court there is a possibility that the lenders lose the judicial process. Also, the process generates a burden to lenders and borrowers represented by the factor J L and J B times the amount of loans, that corresponds to the judicial cost (like lawyers, di erent types of fees, time, etc, J 2 (0; 1)). To follow a peculiar feature of the Brazilian judiciary, the court decides favorably to debtors if their wealth (w 2s + D) is lower than the parameter W 8 : This parameter represents the pro-debtors bias, since the higher is W the bigger is the possibility of loss in court for creditors. It also represents a pro-poor bias, which means that if debtors are richer the chances of creditors win in court is bigger. De nition 1 Strategic default 9 : It occurs when the borrower has enough money to pay his debts (w 2s + D (1 + r)b 0) but he chooses not to do it. De nition 2 Default by bad fortune: It occurs when the realization of states of nature is bad in such way that (w 2s + D (1 + r)b < 0); unabling the borrowers to ful ll their payment promises. The consumption of the rst period de nes the level of debt B at the beginning of period 2: B = (c 1 D w 1 ) which means that the agents consume more than the sum of their wages and durable goods. A loan contract between the borrower and the lender consists of a pair (1 + r; B); where B is the loan volume and (1 + r) the loan rate, subject to the courts costs and its decision that applies to the situation in which the borrower does not repay the debt (1 + r)b. If at least some debt will be held, so that B > 0, we can divide the borrowers actions in four distinct choices: C1 does not le for default if: w 2s + D (1 + r)b and w 2s + D W C2 strategic default if: w 2s + D (1 + r)b and w 2s + D < W C3 bad fortune default (losing in the court) if: w 2s + D < (1 + r)b and w 2s + D W 8 Castelar (2003) presented a careful study about the Brazilian judiciary. The author, interviewing Brazilian magistrates, asked how they would rule in the case of a con ict between compliance with contracts and the interests of less privileged social segments: only 19.7 percent answered that they would follow contracts. 9 Moral hazard is present because borrowers have a choice not to pay their debts. 6

7 C4 bad fortune default (wining in the court) if: w 2s + D < (1 + r)b and w 2s + D < W Notice that C2 and C3 are mutually exclusive. The wealth 8 in each situation for the borrowers is given as follows: w 2s + D (1 + r)b if C1 >< w W 2 = 2s + D J B B if C2 0 if C3 >: [w 2s + D J B B] + if C4 Thus the lender can receive zero in case of default if the court does not seize the borrowers income or a payment between w 2s + D and (1 + r)b otherwise. Analyzing the consumer choice for default, it is optimal if and only if their gains with default are bigger than their gains when they choose not to default: That is, the consumers will default whenever their wealth does not exceed the parameter W ; since they will win in the court (w 2s + D J B B > w 2s +D (1+r)B). This way, we can view the probability of no default as (1 p default ) = X p s s j = p(c1); and the probability s of default as p default = X s p s [ s (1 j ) + (1 s ) j + (1 s )(1 j )] = p(c2) + p(c3) + p(c4), where s = 1 if w 2s + D (1 + r)b and j = 1 if w 2s + D W : Notice that before the new law of payroll loans to be in e ect, the default was chosen if W > w 2s + D; even if the borrowers are totally capable of repaying their debts, i.e. w 2s + D (1 + r)b: With the new law in e ect, this type of default is partially ruled out, since their debts are automatically discounted from their income. Now, the default occurs only if w 2s < (1 + r)b for workers, and never for retired workers and pensioners if w > (1 + r)b 10. For the lenders, the expected return on lending must be no less than the risk-free return. Therefore, the lender s participation constraint is: (1 + r f )B p(c1)(1 + r)b + X s p s (1 s ) j [w 2s + D J B B] + + +[p(c2) + p(c4)]0 [p(c2) + p(c3) + p(c4)]j L B; (1) where p(c1) is the probability of no default, with the borrower paying its debts in full, p(c3) = X p s (1 s ) j is the probability of bad fortune s 10 Notice that we model w 2s as w g s ;thus the repayment is deducted directly from w: Also, since the new law determines that deductions are limited to thirty percent of agents disposable wage, this class of debtors will ever pay their debts. 7

8 default with the lender wining in the court, p(c2) and p(c4) is the probability strategic default and bad fortune default respectively, with lenders loosing in court. The extra interest rate paid r r f is exactly the one needed to o set the loss the nancial institution makes when the consumer defaults: it is the same as an insurance premium. Observe that the lenders expected return, given by the equation (1) determines the supply of credit in the economy. The supply of credit depends directly from the judicial costs (J) and courts pro-debtor bias (W ). Intuitively, if the cost of default is high and the pro-debtor bias is low the borrowers have more incentive in avoid the strategic default. As the judicial costs reduces and the pro-debtor bias increases, the borrowers incentive to le for strategic default raises. Since the court s decision depends on borrowers wealth, the supply of credit may a ect borrowers in di erent ways. For those with higher wealth, the probability of loss in court is higher, inhibiting the strategic default. On the other hand those with lower wealth have higher probability of success in court, motivating the strategic default and increasing the credit constraint. Proposition 1 A pro-debtor judiciary (bigger W ) tends to increase the interest rates. For an extremely high level of bias ( W ) the supply of credit disappears. Proof. For an arbitrary level of W ; the lender recovery all its debt (if w 2s + D (1 + r)b) or at least part of it ( if w 2s + D < (1 + r)b) if s s such that w 2s +D = W : In this case, the probability of payment SP is p s : Let us consider a judiciary with a higher pro-debtors bias, i.e. s=s W > W : Now, just for the states s > s there will be a payment SP P of the debt, reducing the probability of recovery for p s < S p s = s=s s=s SP P p s + s p s. Such increase in W reduces p(c1) and p(c3); and raises s=s s=s p(c2) and p(c4): To keep the equality in the equation (1) r increases. Increasing even more the pro-debtor bias up to the level W that makes the debtor default in all states of nature, the supply of credit disappears since (1 + r f )B = [p(c2) + p(c4)]0 [p(c2) + p(c3) + p(c4)]j L B ) B = 0: Thus, for all W W ; B = 0: Since the people with less wealth are the most bene ted with the judiciary bias, they have an incentive to default in more states than the wealthier people, reducing the expected return of the creditors for this type of agents. When the expected return is lower than the risk-free asset this public have their credit denied. 8

9 Corollary 1 For a given level of pro-debtor bias the poorer people will have the supply of credit denied. Proof. Let W the level of wealth that makes the court decide in favor of the lenders. For the people with goods (D) and income (w 2s ) su ciently low such that w 2s + D < W 8s; the participation constraint (1) will be (1 + r f )B = [p(c2) + p(c4)]0 [p(c2) + p(c3) + p(c4)]j L B and therefore B = 0: The new law of payroll loans provide a kind of protection for the lenders. This law allows creditors to receive their repayment for loans when the debtors income overcome their debt payment, i.e., when w 2s (1 + r)b. This vanish the possibility of strategic default for all states of nature that makes this inequality true. Proposition 2 The new law of payroll loans reduces the interest rates and the credit constraint. Proof. Suppose that W > (1 + r)b; allowing for the possibility of strategic default. Let [s ; s ] the set of states of nature which the agent chooses strategic default, i.e. w 2s + D (1 + r)b and w 2s + D < W ; P making the probability of strategic default equal to s p s : With the new s=s law the possibility of strategic default when w 2s (1 + r)b for all s > s 0 is completely eliminated. In this case, the probability of strategic default P will be s0 P p s < s0 P p s + s p s. Now, the probability of no default increases by s=s s=s s=s 0 P s0 p s ; and p(c2) reduces in the same magnitude.. The new s=s law increases the lenders expected return and to hold the equality in the participation constraint (1) r decreases. The poorest individuals which had their credit denied due to the excessive protection of the court, in all states of nature that w 2s (1+r)B will also pay the debt, increasing the expected return and making (1 + r f )B = p(c2=w 2s (1 + r)b)(1 + r)b [p(c2=w 2s < (1 + r)b)) + p(c3) + p(c4)]j L B ) (1 + r) = (1+r f )+J B [p(c2=w 2s <(1+r)B))+p(C3)+p(C4)] p(c2=w 2s (1+r)B) ; B > 0. The new law has important e ects on the credit market, it reduces the interest rates for all people that uses the mechanism of payroll loans, and spread the credit market to the people that had their credit denied due to the lenience of the local judiciary. The most bene ted with the new law were the poorer and retired people that were excluded from the credit market before the enforcement of law 10,820. With the new law in e ect they started to have access to credit at a low interest rate. 9

10 Corollary 2 The retired people that earn a x income from INSS will have access to credit at the lowest rates, even the poorest ones. Proof. Let w _ be the income received by the retired people. Now, they will automatically ful ll the payment of their promises in all states of nature since their income is x and the uncertainty (g s ) occurs after the payment. Thus, for _ w > (1 + r)b we have (1 + r f )B = (1 + r)b: Remark 1 Notice that we are not considering the probability of death of the retired people. In the real world the interest rates despite smaller, is not equal to the risk-free interest rate, probably because of the risk of death. Remark 2 The other reason of why the interest rate is closer to the riskfree rate is because we are assuming perfect competition between lenders. This might not be the case in the Brazilian credit market. 4 Description of the data and Descriptive Statistics The database comes from Central Bank of Brazil according to circular of 12/30/1999 and comunicado of 05/20/2000. The database contains daily information about new loans, total amount of credit and interest rates. All information is displayed by nancial institution, type of loan indexation (pre xed, interest rate indexation, exchange rate indexation and price indexation) and kind of borrower (household or rm). The present paper uses information about personal loans, i.e., credit to households, without collateral but with considerable screening by - nancial institutions. The loan is not linked with the purchase of a speci c good, so the borrower can buy any kind of good or even repay other debts with the loans proceeds. We use only the pre xed loans, since it is the most usual type of personal credit 11. The sample consists in a group of 112 nancial institutions, which contains a subsample of 40 nancial institutions that are able to give personal loan according to the new law. We run our empirical tests using monthly data generated through the daily data. The new loans variable is constructed by summing of 11 The personal loans are composed basically of two kinds of personal credit: direct credit to consumer (CDC) and payroll loans (Crédito consignado). The basic di erence between them is that payroll loans allow the direct discount of the debt payment on the payroll check while direct credit to consumer doesn t. 10

11 daily values over the whole month. The total amount of credit is de ned as the monthly average of the daily data for the stock of loans. Finally, for the interest rates variable, we use the monthly average, weighting up by respective new loan volume. The sample period is from January of 2003 to October of 2005, so our sample consists of 34 months and 112 cross sections units (3,808 observations). Now we present some descriptive statistics concerning personal loans for treatment and control groups before and after the new law of payroll loans take e ect. Figure 3 illustrates a strong increase of the new loans volume for the treatment group in the post-treatment period. The monthly average of new loans increased from R$57 millions to R$127 millions, i.e., a rise of R$70 millions. The coe cient of variation (standard deviation/mean) decreased from de 2.7 to 1.9, meaning a fall of the new loans dispersion for the treatment group in the post treatment period. The same cannot be told about the control group (see table 1). Comparing with the treatment group, the increase in the mean of new loans for the control group was not so high, raising from R$25 millions to R$35 millions, i.e. a rise of R$10 millions. Also, the coe cient of variation practically did not change. Figure 3: Mean of new loans before and after the new law 150 pre treatment post treatment New loans (millions) treatment Group control Table 1: New Loans Treatment Group Control Group pre treatment post treatment pre treatment post treatment mean standard deviation std/mean 2,7 1,9 2,8 2,7 The monthly average interest rate (% per month) for the treatment group decreased from 5.7% before treatment to 4.3% in the post treat- 11

12 ment period, i.e., a reduction of 1.4% per month, which annualized means a fall of 18% per year. The variation coe cient decreased from 0.56 to 0.49, meaning that the interest rate standard deviation also decreases (see table 2). For the group of control, the monthly average interest rate decreased from 5.5 % to 4.8%, which means that the fall for the treatment group was two times bigger than the fall for the control group, 1.4% against 0.7% (see gure 4). Even though there were timing factors that decreased the interest rates for both control and treatment groups, the reduction for the treatment group was much higher. The variation coe cient for the control group remained stable. Figure 4: Mean of interest rate before and after the new law Interest rate (month) 6,0% 5,0% 4,0% 3,0% 2,0% pre treatment treatment Group post treatment control Table 2: Interest Rate Treatment Group Control Group pre treatment post treatment pre treatment post treatment mean 5,7% 4,3% 5,5% 4,8% standard deviation 3,2% 2,1% 3,0% 2,5% std/mean 0,56 0,49 0,54 0,52 Figure 5 illustrates the higher increase of total amount of personal credit for the treatment group when compared with the control group. However, such di erence is not as larger as before, which can be explained by the duration of the contracts. Usually, such contracts have duration varying between 1 to 3 years, which means that a signi cant part of the total amount is prior to the new law. Thus, since the new law is recent, is natural to expect that the e ect on total amount of credit be smaller than the e ect on new loans. The monthly average of total amount of credit increased from R$475 millions to R$993 millions. The coe cient of variation (standard deviation/mean) decreases from de 2.8 to 2.2, reducing the dispersion of the amount of credit for the treatment group in the post treatment period (see table 3). For the control group, the total amount of credit increases from R$187 millions to R$342 millions. There was no change in the coe cient of variation. 12

13 Figure 5: Mean of the total amount of credit before and after the new law 1200 pre treatment post treatment Total amount of credit (millions) treatment control Group Table 3: Total Amount of Credit Treatment Group Control Group pre treatment post treatment pre treatment post treatment mean standard deviation std/mean 2,8 2,2 2,8 2,8 Analyzing the above information as growth rate, we notice that the new law a ects strongly the treatment group in all considered variables (see gure 6 and table 4). First, the new loans variable su ered a much higher increase in the treatment group than in the control group, 123% against 42%. Second, despite the law be recent, the total amount of credit for the treatment group also su ers a signi cant impact after the new law be in force, 109% against 83%. Finally, the reduction of the interest rate was much bigger for treatment group too, 23% against 13%. Figure 6: Means Growth rate - Treatment x Control 150% 100% 50% Treatment Control 0% 50% New loans interest rate total amount 13

14 Table 4: Mean's growth rate treatment x control New loans interest rate total amount Treatment 123% 23% 109% Control 42% 13% 83% To sum up, the descriptive statistics show that between pre and post treatment periods, there was a generalized downward trend in the interest rate and upward trend in the new loans volume and total amount of credit, but these trends were much stronger within treatment group. 5 Empirical Tests We measure the impact of the new law of payroll loans on personal credit market (new loans, interest rate and total amount of credit) using a di erence-in-di erence estimator 12 with bank-level panel data. The di erence-indi erence speci cation compares the change (before and after the government implements the new law) in personal loans in treatment banks (those authorized by the government to supply such type of loan) with the corresponding change in control banks (those without permission to o er such loan). Unlike the usual di erence-in-di erence procedures, the pre and post treatment period will be di erent for each institution because each institution became able to give payroll loan according to the new law in di erent periods. For the control group we assumed May of as a cut for pre and post periods. The estimation was made using several empirical models to access the new law s e ect on new loans volume, interest rate and total amount of credit. For all three dependent variables we estimate: a di -in-di model without xed e ects; a di -in-di model with xed e ects by cross section unit; and a di -in-di model with cross section unit and time xed e ects. Since the variable new loans is censored in zero, in this case we also estimated a Tobit model. The empirical model OLS1 can be speci ed in regression form as: y it = c + dum_treatment i + dum_period it + + dum_treatment i dum_period it + " it : 12 See Meyer (1995). 13 This was the date that the rst institution became able to give payroll loan according to the new law. To test the robustness of our results we ran our models using April 2005 the date which more than 50% of the treatment group was able to give payroll loans as a cut for pre and post period for the control group. The results did not change in terms of coe cient magnitudes and signi cance. 14

15 The right-hand-side of OLS1 includes dum_treatment i to control for speci c factors of treatment group xed over time; the dum_period it to control for factors that vary from the period pre to post treatment but are common across all banks; and the interaction of both dummies that is the di -in-di variable that captures the impact of new law of payroll loans. The empirical model OLS2 can be speci ed in regression form as: y it = c i + dum_period it + dum_treatment i dum_period it + " it : The right-hand-side of OLS2 includes the bank xed e ects to control for bank-speci c factors that are xed over time, the dum_period it to control for factors that vary from the period pre to post treatment but are common across all banks and the interaction of both dummies that is the di erence-in-di erence variable that measures of the impact of new law of payroll loans. The empirical model OLS3 can be speci ed in regression form as: XT 1 y it = c i + ' t period t + dum_treatment i dum_period it + " it : t=1 For the dependent variable new loans the empirical form used to specify the Tobit model was: XT 1 NL it = c i + ' t period t + dum_treatment i dum_period it + " it t=1 NL it = NL it if NL it > 0; NL it = 0 otherwise. Finally, the right-hand side of OLS3 and Tobit include bank- xed e ects to control for bank-speci c factors that are xed over time, the month xed e ects to control for factors that vary over time but are common across all banks, and the coe cient that is the di erence-indi erence estimator of the impact of new law of payroll loans. We de ne the dummies variables as follows: dum_treatment i is a dummy variable that assumes 1 when the cross section unit i belongs to the treatment group (the institutions able to give payroll loans according to the new law) and 0 otherwise. dum_period it is a dummy variable that assumes 1 if the observed period for cross section unit i is after its inclusion in the payroll loans 15

16 concession according to the new law and 0 otherwise. For the institutions that are not able to give payroll loans this dummy variable assumes 0 before May of 2004 and 1 otherwise. May of 2004 is the date that the rst institution became able to give payroll loans according to the new law. 5.1 New Loans In this subsection we estimate the e ect of the new law of payroll loans on new loans. The coe cients estimated from the four di erent models are presented at table 5. The key variable is the interaction of an indicator of whether the bank is able to o er such loan with indicator of whether the month was in the post treatment period. In other words, the main interest of the estimation is to access the economic and statistical signi cance of the interaction between dum_treatment i and dum_period it. It provides the real e ect of treatment (the new law) on the variable of interest. The estimated e ect of the new law on new loans is statistically signi cant at 1% level in all speci cations and its magnitude varies between R$41.8 millions and R$56.7 millions. As we showed previously, the average of new loans for the treatment group before the new law was R$57 millions with an increase of R$70 millions after the new law. Thus, the e ect caused by the new law on the new loans varies between 73% and 100% (between 60% and 81% of the total increase). Table 5: new loans (R$ millions)* OLS1 OLS2 OLS3 Tobit constant (0.59) dum_treatment 32.1 dum_period (0.04) dum_treatment x dum_period time fixed effects No No Yes Yes cross section fixed effects No Yes Yes Yes R 2 0,05 0,91 0,91 0,09 Number of cross sections Number of obs * p value in parentheses Robust standard errors (except the Tobit model) 16

17 These results show the potential improvements in social welfare that this law may produce. Such law helps to complete the contingent markets once it brings the possibility of people making contracts of future payments conditioned to future cash ow. It reduces the number of states of nature in which banks would lost their money in a pro debtor judicial decision, stimulating the supply of credit (see proposition 2) mainly to people who did not have enough personal goods to use as collateral for borrowing from bank sector. All this factors make the volume of new loans increase, specially for the retired workers from the INSS (see corollary 2), which were the biggest share of population a ected by the new law. Using the di -in-di methodology we are able to take out the bias of a "pure estimator". We showed in other section that the new loans volume increased for both treatment and control group between pre and post treatment periods. So it is important to control for this generalized trend of growth in the credit market through the use of di -in-di estimator. The results show that even controlling for common temporal trends and for cross section heterogeneity (the time and cross section xed e ects) the new law s e ect on new loans is economic and statistical signi cant. It means the new loans growth between pre and post treatment periods was much bigger for treatment group than for control group even when we control for these factors. 5.2 Interest Rate Table 6 presents the main results for the dependent variable interest rate. The di erence-in-di erence coe cient is negative and statistical significant at 5% in all estimations. This result indicates that the new law caused a reduction in the interest rate on personal loans. The estimated decrease varies between 0.24% per month to 0.61% per month, which means a decrease varying between 2.9% and 7.6% per year. Comparing to the levels of interest rate for the treatment group before treatment period we can say that the rate decreased between 4% and 11% (between 17% and 44% of the total fall). 17

18 Table 6: interest rate (percent)* OLS1 OLS2 OLS3 constant dum_treatment 0.17 (0.16) dum_period dum_treatment x dum_period (0.02) 0.31 time fixed effects No No Yes cross section fixed effects No Yes Yes R 2 0,02 0,87 0,89 Number of cross sections Number of obs * p value in parentheses Robust standard errors The model of the previous section rationalizes this e ect by the supply side of credit market. With the new law the expected value recovered by the bank is bigger since it will be repaid in more states of nature, once the new law vanishes a signi cative part of the default, which occurs only when the debtor loose his job or dies. So, the desired interest rate for the bank that equalizes the expected return of the credit and the risk free interest rate reduces (see proposition 2). We note that while the new loans volume raised between 73% and 100% the interest rate dropped only between 4% and 11%. Why didn t the interest rate drop more deeply? One possible explanation is that our tests do not consider the people who tried to take loans but the extremely higher interest rates makes them turned down the credit. Thus, our result would be a lower bound of such fall in the interest rate. Another possible explanation is that after the new law the personal loans demand increased putting upward pressure on interest rate. Besides, probably most of potential borrowers of payroll loans were rationed before the new law. When there is rationing in credit markets, due to non monotonic credit supply function, the quantity moves while interest rate remains xed after a supply shock Total Amount of Credit In table 7 the variables are de ned the same way as before. The estimated e ect of the new law on total amount of credit was positive and statistical signi cant at 1%. The e ect was estimated between R$239.3 millions and R$363.7 millions. Comparing to the volume of credit for 14 See Stiglitz and Weiss (1981), corollary 1 pp

19 the treatment group before the treatment period where the amount of credit was R$445 millions, we can say that the estimated growth of the total amount of credit varies between 50% and 76%. Also, since the increase was R$518 millions, the e ect of the new law corresponds to 46% to 70% of the total increase. Thus, the new law s e ect was not only statistical but also economic signi cant. This nding con rms the previously reported results for new loans. Table 7: Total amount of credit (R$ millions)* OLS1 OLS2 OLS3 constant dum_treatment (0.16) dum_period dum_treatment x dum_period (0.01) time fixed effects No No Yes cross section fixed effects No Yes Yes R 2 0,04 0,91 0,91 Number of cross sections Number of obs * p value in parentheses Robust standard errors 6 Conclusion The new Brazilian law of payroll loans, launched in September 2003, provides a natural experiment on personal credit, since it was an exogenous change that a ected only a fraction of nancial institutions. Di erently from the past where only workers, pensioners and retires from the public sector had access to payroll loans, the new legislation regulates such loan to private workers associated to trade unions, pensioners and retired workers from INSS. This type of credit allows nancial institutions to deduct the monthly repayment of debt compulsorily, eliminating a signi cant part of the strategic default. Theoretically, banks will be more willing to o er better terms of credit since the expected value of the loan s repayment increases due to the lower probability of default. The empirical methodology applied to estimate the e ect of the new law was the di erence-in-di erence procedure, using the institutions authorized by the government to o er payroll loans as the treatment group. The results that are totally aligned with our theoretical claims point 19

20 to an increase in the new loans and in the total amount of credit, and to a fall in the interest rate. In quantitative terms, our estimation indicates an increase in the new loans varying between 73% and 100% and in the total amount of credit varying between 50% and 76%. For the interest rate there was found a decrease that varies between 4% and 11%. All results are highly signi cant and robust to the empirical model. References [1] Banerjee, Abhijit V., Gertler, Paul J. and Ghatak, Maitreesh "Empowerment and e ciency: tenancy reform in West Bengal", Journal of Political Economy, 2002, vol.110, no.2. [2] Castelar, Armando Judiciário, reforma e economia: A visão dos magistrados. Working paper 966. Rio de Janeiro: Institute of Applied Economics Research, [3] Coelho, Christiano A., Garcia, Marcio and Mello, João M. "Testing for credit rationing in Brazilian market", Mimiographed, PUC-Rio, [4] Costa, Ana C. and Mello, João M. "Judicial Risk and Creditor Expropriation: Micro Evidence from Brazilian Payroll Loans", PUC- Rio Workshop, [5] Meyer, Bruce D. "Natural and quasi-experiments in economics", Journal of Business and Economics Statistics, 13 (2): , [6] Stiglitz, Joseph E., Weiss, Andrew, "Credit rationing in markets with perfect information", American Economic Review, vol. 71, no.3,

21 A Appendix Table A1 Number of institutions signing the deal with INSS by month May of September of November of December of January of March of April of May of June of August of Source: Homepage of national press (diário oficial): 21